ap-wait-durable

Wait For Sessions Without Blocking (Durable)

When to use

You need to know when a spawned agentproto session finishes, but you don't want to sit inside a single turn blocking on it — either because the wait could be long, or because you want to keep working on other things in the meantime. This is the non-blocking counterpart to ap-wait-fanin.

Pattern: background CLI wait with a notification

Run the wait as a background terminal process and let its completion notify you, instead of parking your own turn on it.

agentproto sessions wait sess_abc123 --until turn-end

Launched via a Hermes terminal with background=true, notify_on_complete=true, the wait runs off to the side and its completion arrives as a new message — zero polling inside your own context.

agentproto sessions --watch

Alternative for open-ended supervision of many sessions at once rather than one specific wait.

Pattern: cron re-ping as a durable fallback

For waits that must survive a daemon restart, or where a background terminal isn't an option, arm a cron job that re-pings your own living session by id.

MCP: cron_create

{
  "tool": "cron_create",
  "args": {
    "schedule": "*/10 * * * *",
    "action": { "kind": "prompt-session", "sessionId": "sess_self001" },
    "prompt": "Check whether sess_abc123 has finished its turn yet."
  }
}

MCP: cron_list / cron_run / cron_delete

{ "tool": "cron_list", "args": {} }

{ "tool": "cron_run", "args": { "cronId": "cron_77f1" } }

{ "tool": "cron_delete", "args": { "cronId": "cron_77f1" } }

cron_run fires the job immediately (useful to test it before trusting the schedule); cron_delete disarms it once the wait is over — a stale re-ping cron will keep nagging a session that no longer needs it.

Gotchas

• From a cowork/Hermes session, prefer terminal(background=true, notify_on_complete=true) over anything foreground — the notification lands as a new message with zero polling cost to your context.
• cron_create with prompt-session re-pings a LIVING session by id. It is the durable alarm that survives daemon restarts, but it is a nudge that re-invokes you to go check — not a wait result delivered automatically. Your re-invoked turn still has to call agent_output or session_events_poll to actually learn the outcome.
• Never foreground-block your own turn with a long sleep to simulate a wait — watchdogs will kill long in-turn sleeps, and (per ap-wait-fanin) killing a wait mid-flight can corrupt the session being waited on.
• Delete the cron once the wait resolves; an armed re-ping left behind will keep firing on schedule and burning tokens on a session that's long done.